I’ve been trying to integrate a newly created installation of Ollama with Home Assistant, and get Assist to be a bit smarter. I have set up the Ollama Integration, which is pointing to my Ollama server on my Unraid machine. No issues getting it to connect and see the various Models I have downloaded.
However, none of the models seem to actually work in terms of getting good information or turning a light on/off. I read that llama3.2 seems to be a good option, and have added that to my Ollama server.
However, all I get when sending in any prompt is “Unexpected error during intent recognition”.
Is there some kind of configuration I missed specific to llama3.2? Is anyone using it currently and could perhaps provide a few pointers?
Your log should give a clue as to why. Look for the log text in the core log.
Second is this the first time you tried to setup an LLM?
Third you will also find the voice debug log under settings, voice assistants. This is different than the core log which will tell you what blew up. This log should tell you what was sent to be blown up.
Thank you! I figured it out. For some reason, the logs weren’t showing the right error last night, but I tried again this morning to get a fresh entry and finally saw that the model didn’t support “Thinking” (which I find funny). I changed it and now it works!
Response time on my server isn’t too bad either, it’s hosted on an Unraid server built with 2x E5-2630 v2s and an RTX 2080. I was setting this up in preparation for getting a Voice assistant one day, so wanted to see what a fully local setup would look like.